Automated Code Review in Practice
Umut Cihan, Vahid Haratian, Arda İçöz, Mert Kaan Gül, Ömercan Devran, Emircan Furkan Bayendur, Baykal Mehmet Uçar, Eray Tüzün
Abstract
Context: Code review is a widespread practice among practitioners to improve software quality and transfer knowledge. It is often perceived as time-consuming due to the need for manual effort and potential delays in the development process. Several AI-assisted code review tools (Qodo, GitHub Copilot, Coderabbit, etc.) provide automated code reviews using large language models (LLMs). The overall effects of such tools in the industry setting are yet to be examined. Objective: This study examines the impact of LLM-based automated code review tools in an industry setting. Method: The study was conducted within an industrial software development environment that adopted an AI-assisted code review tool (based on open-source Qodo PR Agent). 238 practitioners across ten projects had access to the tool. We focused our analysis on three projects, which included <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{4, 3 3 5}$</tex> pull requests, 1,568 of which underwent automated reviews. Our data collection comprised three sources: (1) a quantitative analysis of pull request data, including comment labels indicating whether developers acted on the automated comments, (2) surveys sent to developers regarding their experience with the reviews on individual pull requests, and (3) a broader survey of 22 practitioners capturing their general opinions on automated code reviews. Results: <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$73.8\%$</tex> of automated code review comments were labeled as resolved. However, the overall average pull request closure duration increased from five hours 52 minutes to eight hours 20 minutes, with varying trends observed across different projects. According to survey responses, most practitioners observed a minor improvement in code quality as a result of automated code reviews. Conclusion: The LLM-based automated code review tool proved useful in software development, enhancing bug detection, increasing awareness of code quality, and promoting best practices. However, it also led to longer pull request closure times and introduced drawbacks such as faulty reviews, unnecessary corrections, and irrelevant comments. Based on these findings, we discussed how practitioners can more effectively utilize automated code review technologies.